The current solution is overly complex and as such not maintainable.
Further we have known defects such as missing handling of embedded messages, as well as large swaths of code that really shouldn't be necessary (the whole temporary file handling madness). Also, the amount of ways we transform the same content over and over and dependency injection isn't going to help in any future bugfixes.

The aim is to generate a maintainable solution that suits our needs.
It will focus on querying rather than walking the tree and dealing the content in a visitor pattern.